Thermal NO Predictions in Glass Furnaces: A Subgrid Scale Validation Study

Transcription

1 Feb 12 th 2004 Thermal NO Predictions in Glass Furnaces: A Subgrid Scale Validation Study Padmabhushana R. Desam & Prof. Philip J. Smith CRSIM, University of Utah Salt lake city, UT th Annual ACERC Conference Provo, UT

2 Glass furnaces: Thermal NO High pre-heated air is used in the combustion for fuel efficiency, results in high peak flame temperatures (typically around 2200 K) Thermal NO formation is very significant above 1800 K Forms in local regions, where temperature is high & radicals such as O, OH present To meet the environmental regulations, glass manufacturers are in need of cost-effective tools to minimize the emissions Photograph from the Society of Glass Technology Combustion in a glass furnace Cross-sectional view of a single port

3 Turbulent mixing & reaction-nox (Nakamura, Smart, and Van de Kamp, J. Inst. Energy 69, 1996, 39-50)

4 Time & Length scales Mesh/subgrid

5 Numerical simulation of combustion Direct Numerical Simulations (DNS) is not possible for practical problems in the foreseeable future Resolves all the scales, both spatially and temporally Large Eddy Simulations (LES) is difficult, but possible Resolves problem-dependent large scales, models small scales Reynolds Average Navier Stokes (RANS) simulations is the feasible solution for industrial scale problems Solves time-averaged governing equations Needs subgrid scale models to account for the unresolved scales Turbulence model: Accounts for the unresolved turbulent scales on the mean flow transport Mixing model: Represents mixing at subgrid scales Reaction model: To simplify the complex finite-rate calculations

6 Reaction model (Thermo chemistry) Reduces the number of degrees of freedom associated with the combustion chemical reactions in CFD calculations If the state of the system (ø i ) has n+2 degrees of freedom i.e., [ ρ, T, Y1,... Y n ],a reaction model parameterizes the state with one/more independent tractable variables Integration of stiff PDEs can be avoided in CFD For non-premixed combustion, two widely used models Equilibrium model: Mixture fraction (f) Steady flamelets model : Mixture fraction (f) & scalar dissipation (χ) An important underlying assumption is that mixing is the rate limiting process compared to chemical reactions (High Damkholer number)

7 Reaction model (continued) φ i = R( f ) Equilibrium φ i = R( f, χ) Laminar flamelets

8 Thermal NO chemistry Extended Zeldovich mechanism O + N 2 N + O 2 k + 1 NO + N k + 2 NO + O k + 3 N + OH NO + H With the quasi-steady state assumption for N atoms d[ NO] dt 2 k k NO 1 2[ ] 1 k N k O k O N 1[ 2 ] 2[ 2 ] = 2 1 [ ][ 2 ] gmol / m k NO 1[ ] 1 + k O k OH 2[ 2 ] + 3[ ] 3 s How should we choose the intermediates O & OH? Equilibrium Partial equilibrium Instantaneous quantities from advanced subgrid reaction models (nonequilibrium effects) Turbulence-NO chemistry effects? Needs a mixing model Which NO should we select?

9 Nonequilibrium effects: O & OH Mass fraction of O Mass fraction of OH

10 Nonequilibrium effects: O & OH Mass fraction of O Mass fraction of OH

11 DNS validation of reaction models * DNS Data Equilibrium DNS Data Flamelets DNS of spatially evolving non-premixed CO-H 2 jet Compared to equilibrium chemistry models, flamelets predicted the OH concentrations reasonably *Courtesy of James Sutherland, CRSIM

12 Mixing model Accounts for mixing at unresolved (subgrid) scales Subgrid scale statistics can be represented with a prescribed probability density function from moments of the tractable variables computed on the mesh Equilibrium Resolving the moments at grid level is crucial to represent the subgrid scale mixing accurately φi = φi ( f ) P( f ) df Laminar flamelets Low grid resolution φi = φi ( f, χ) P( f, χ) dfdχ Mixing & State space High grid resolution Resolution &mixing

13 Complete System Prediction/Minimization of NOx emissions from glass furnaces Combustion Space Glass Melt IFRF Glass Furnace Single Port Model Two Port Model TNF Flame Laminar CFD Turbulence Model Mixing Model Reaction Model NOx Model Radiation Model Soot Model

14 IFRF glass furnace Furnace Dimensions: 3.8 m long X 0.88 m wide X m high Grid: Modeled only half the domain 420,000 hexahedral elements (after grid adaption) Validation data* on the plane of symmetry at x=0.6, 0.9, 1.2, 1.8 & 2.4 m (along the vertical direction) Temperature, O 2,CO 2,CO,CH 4,NOx Operating conditions Natural gas at 283 K 10 % excess air at 1373 K Outlet Air inlet Fuel inlet T.Nakamura, W.L. Vandecamp and J.P. Smart, Further studies on high temperature * gas combustion in glass furnaces, IFRF Doc No F 90/Y/7, August 1991

15 Simulation details Parallel version of FLUENT 6.0 on 4 processors Flow & Turbulence: Time-averaged Navier-Stokes equations with standard κ-ε model & RSM for turbulence closure Combustion: Mixture fraction with equilibrium chemistry & flamelets Flamelets: GRI Mech2.11* chemical mechanism Radiation: Discrete-ordinates with weighted-sum-of-graygases model (WSGGM) for gas absorption coefficients Soot: Two-step Tesner model (soot formation & combustion) with participation in radiation Boundary conditions: Velocity is specified at the fuel and air inlets Wall B.C.s & glass surface are treated by specifying heat flux * C.T. Bowman, R.K. Hanson, D.F. Davidson, W.C. Gardiner, Jr., V. Lissianski, G.P. Smith, D.M. Golden, M. Frenklach and M. Goldenberg,

16 Temperature Distribution 0.9m 1.8m 0.6m 1.2m Temperature( 0 C) contours in the plane of symmetry

17 Temperature validation Distance from bottom (m) K-epsilon,Equilibrium K-epsilon,Flamelets RSM,Equilibrium RSM,Flamelets Experiment Distance from bottom (m) K-epsilon,Equilibrium K-epsilon,Flamelets RSM,Equilibrium RSM,Flamelets Experiment Temperature (C) X=0.6 m Temperature (C) X=0.9 m

18 O 2 validation K-epsilon,Equilibrium K-epsilon,Equilibrium Distance from bottom (m) K-epsilon,Flamelets RSM,Equilibrium RSM,Flamelets Experiment Distance from bottom (m) K-epsilon,Flamelets RSM,Equilibrium RSM,Flamelets Experiment Mole fraction of O 2 X=0.6 m Mole fraction of O 2 X=1.8 m

19 NOx calculations Turbulence: Sensitivity of turbulence model is studied with standard κ-ε model and RSM Mixing: Turbulence effects on the NOx production rates are accounted through the mixture fraction PDF Reaction model: NOx is post-processed with the following O & OH radical concentrations In the case of equilibrium combustion calculations, O & OH are taken from the partial-equilibrium approximation For flamelets combustion calculations, O & OH concentrations are from flamelets PDF look-up tables

20 NOx validation K-epsilon,Equilibrium K-epsilon,Equilibrium K-epsilon,Flamelets K-epsilon,Flamelets 0.80 RSM,Equilibrium RSM,Flamelets 0.80 RSM,Equilibrium RSM,Flamelets Experiment Experiment Distance from bottom (m) Distance from bottom (m) NOx (ppm) X=0.6 m NOx (ppm) X=0.9 m

21 NOx validation 1.00 K-epsilon,Equilibrium 1.00 K-epsilon,Equilibrium Distance from bottom (m) K-epsilon,Flamelets RSM,Equilibrium RSM,Flamelets Experiment Distance from bottom (m) K-epsilon,Flamelets RSM,Equilibrium RSM,Flamelets Experiment NOx (ppm) NOx (ppm) X=1.2 m X=1.8 m

22 Large Eddy Simulations (LES) NS equations are filtered to retain large scales of the flow Large scales are more problem-dependent and contains most of the energy Needs subgrid scale models for small scales, which tends to have more universal behavior Resolves flow and mixing more accurately than RANS methods LES of a TNF workshop flame

23 LES & thermal NO Temperature & Thermal NO source

24 Conclusions Mixing: Resolving mixing is crucial in predicting the local thermo-chemical state of the system and pollutants Resolved scale mixing: Predictions are very sensitive to the inlet boundary profiles Subgrid scale mixing: LES resolves mixing more accurately than RANS, thus reduces the burden on mixing model Reaction Model: For NOx predictions, the intermediate species should be chosen from realistic reaction models, which can include the nonequilibrium effects Validation: A systematic validation strategy for NOx simulation in industrial furnaces needs to include validation at pilot and bench scales.

25 Acknowledgement This research program is financially supported by Sandia National Laboratories & DOE Office of Industrial Technologies, under the supervision of Dr. Robert J. Gallagher